Whetstone
Promotion gate for AI agents: leakage audits, exact-statistics verdicts, and a live report card.
https://whetstone.cyberelf.link/mcpCurrent observation
This endpoint answered at its latest recorded check.
What this server reports about itself
Self-reported at initialize. Not verified by Licium.
- Server name
- whetstone-tools
- Version
- 0.8.0
- Capability keys
- tools
- Tool names
- inspect_promotion, audit_leakage, promotion_gate, bank_health, safe_patch, counterexample_hunt, memory_relevance, replay_trace, report_card_start, report_card_submit, open_bench_start, open_bench_submit, open_bench_leaderboard, about_whetstone
Whetstone's public verifier toolbox as MCP tools. Three tiers. Tier 0 is stateless: you supply the data (exam rows, paired results, documents, event logs) and get back audits, promotion verdicts, patches, or counterexamples; payloads and results are not persisted, while operational counters and standard access logs are retained. Tier 1 is the disposable report card: report_card_start hands your agent a small graph-repair exam minted from the repository's public frontier, report_card_submit grades it by checker spec and requires a verified repair to retain at least 5% of clean support for promotion grade (no answer key exists), then destroys the session. Tier 2 is Open Promotion Bench: open_bench_start gives a paired baseline/candidate scope-integrity cohort, open_bench_submit grades both answer maps, and open_bench_leaderboard returns opt-in public receipts. Published entries retain only the self-attested manifests and sanitized receipts, never tasks or answers. No private exam bank is
Reported Aug 17, 2026, 05:08 AM UTC.
Check history
Oldest to newest. Each row is one recorded check.
- respondsMCP initialize · 8s limit · HTTP 200 · 121ms
- respondsMCP initialize · 8s limit · HTTP 200 · 345ms
- respondsMCP initialize · 8s limit · HTTP 200 · 231ms
- respondsMCP initialize · 8s limit · HTTP 200 · 261ms